Lectures 10 -11

Repeated Games

14.12 Game Theory

Muhamet Yildiz

Road Map
1.

Forward Induction Examples

2. Finitely Repeated Games with observable actions

1. Entry-Deterrence/Chain-store paradox
2. Repeated Prisoners Dilemma
3. A general result
4. When there are multiple equilibria

3. Infinitely repeated games with observable actions

1.
2.
3.
4.
5.

Discounting / Present value

Examples
The Folk Theorem
Repeated Prisoners Dilemma, revisited tit for tat
Repeated Cournot oligopoly

Forward Induction
Strong belief in rationality: At any history
of the game, each agent is assumed to be
rational if possible. (That is, if there are two
strategies s and s of a player i that are
consistent with a history of play, and if s is
strictly dominated but s is not, at this
history no player j believes that i plays s.)

Burning Money
0

B
3,1

S
.1,.1

S .1,.1

1,3

S -.9,.1

BB

BS

SB

SS

3,1

3,1

.1,.1

.1,.1

.1,.1

1,3

1,3

DB 2,1

-.9,.1

2,1

-.9,.1

DS -.9,.1

0,3

-.9,.1 0,3

O T

B
2,1

S
-.9,.1
0,3

0B
0S

.1,.1

Repeated Games

Entry deterrence
1
X

Enter

2 Acc.

(1,1)

Fight
(0,2)

(-1,-1)

Entry deterrence, repeated twice,

many times
1
X
(1,3)

Acc.

2 Enter

Fight
(-1,1)

(0,4)

Enter

2 Acc.

Fight

1
X

2 Acc.

Enter

Fight

(1,3)
(0,0)
Enter 2 Acc.

(2,2)

(0,0)

Fight
(-2,-2)

(-1,1)

What would happen if repeated n times?

Prisoners Dilemma, repeated twice,

many times
Two dates T = {0,1};
At each date the prisoners dilemma is played:

C 5,5

0,6

D 6,0

1,1

At the beginning of 1 players observe the strategies at 0.

Payoffs= sum of stage payoffs.

Twice-repeated PD
1

D
2

C
1

C
10
10

1
D
2

1
D

D C

D C

5
11

6
6

0
12

5
11

D C
11
5

6
6

D
2

2
D

1
7

C
11
5

D C

6
6

12
0

7
1

6
6

1
7

2
2

7
1

What would happen if T = {0,1,2,,n}?

A general result
G = stage game = a finite game
T = {0,1,,n}
At each t in T, G is played, and players remember
which actions taken before t;
Payoffs = Sum of payoffs in the stage game.
Call this game G(T).
Theorem: If G has a unique subgame-perfect
equilibrium s*, G(T) has a unique subgameperfect equilibrium, in which s* is played at each
stage.

With multiple equilibria

T = {0,1}
2

M2

1,1

5,0

0,0

M1

0,5

4,4

0,0

0,0

0,0

3,3

Infinitely repeated Games with

observable actions
T = {0,1,2,,t,}
G = stage game = a finite game
At each t in T, G is played, and players
remember which actions taken before t;
Payoffs = Discounted sum of payoffs in the
stage game.
Call this game G(T).

Definitions
The Present Value of a given payoff stream =
(0,1,,t,) is
PV(;) = t=1 tt = 0 + 1 + + tt +
The Average Value of a given payoff stream is
(1)PV(;) = (1)t=1 tt
The Present Value of a given payoff stream at t is
PVt(;) = s=t s-t s = t + t+1 + + st+s +

Infinite-period entry deterrence

1
X
(0,2)

Enter

2 Acc.
Fight
(-1,-1)

(1,1)

Strategy of Entrant:
Enter iff
Accomodated before.
Strategy of Incumbent:
Accommodate iff
accomodated before.

Incumbent:
V(Acc.) = VA =
V(Fight) = VF =
Case 1: Accommodated before.

Entrant:
Accommodated
Enter =>
X =>

Not Acc.

Fight =>
Acc. =>

Enter =>
X =>

Case 2: Not Accommodated

Fight =>
Acc. =>
Fight

Infinitely-repeated PD
C

C 5,5

0,6

D 6,0

1,1

A Grimm Strategy:
Defect iff someone
defected before.

VD = 1/(1);
VC = 5/(1) = 5VD;
Defected before (easy)
Not defected
D =>
C =>
C

Tit for Tat

Start with C; thereafter, play what the other
player played in the previous round.
Is (Tit-for-tat,Tit-for-tat) a SPE?
Modified: Start with C; if any player plays
D when the previous play is (C,C), play D
in the next period, then switch back to C.

Folk Theorem
Definition: A payoff vector v = (v1,v2,,vn) is feasible
iff v is a convex combination of some pure-strategy
payoff-vectors, i.e.,
v = p1u(a1) + p2u(a2) ++ pku(ak),
where p1 + p2 ++ pk = 1, and u(aj) is the payoff
vector at strategy profile aj of the stage game.
Theorem: Let x = (x1,x2,,xn) be s feasible payoff
vector, and e = (e1,e2,,en) be a payoff vector at
some equilibrium of the stage game such that xi > ei
for each i. Then, there exist < 1 and a strategy
profile s such that s yields x as the expected
average-payoff vector and is a SPE whenever > .

Folk Theorem in PD
C

C 5,5

0,6

D 6,0

1,1

A SPE with PV
(1.1,1.1)?
With PV (1.1,5)?
With PV (6,0)?
With PV (5.9,0.1)?

Infinitely-repeated
Cournot oligopoly
N firms, MC = 0; P = max{1-Q,0};
Strategy: Each is to produce q = 1/(2n); if any
firm defects produce q = 1/(1+n) forever.
VC =
VD =
V(D|C) =
Equilibrium

10

1
0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5

20

40

60

80

200

400

600

800

100

1
0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5

1000

11

IRCD (n=2)
Strategy: Each firm is to produce q*; if any one
deviates, each produce 1/(n+1) thereafter.
VC = q*(1-2q*)/(1-);
VD = 1/(9(1-));

2
VD|C = max q(1-q*-q) +VD = (1 q *) / 4 +
9(1 )
Equilibrium iff
2
q * (1 2q *) (1 )(1 q *) / 4 + / 9

q*

9 5
3(9 )

x = , y = (3-5/3 )/(9- )

0.4
0.3

0.2
0.1
0
0

0.2

0.4

0.6

0.8

12

Carrot and Stick

Produce at the beginning; at ant t > 0, produce if both
produced or both produced x at t-1; otherwise, produce
x.
Two Phase: Cartel & Punishment
VC = 1/8(1-). Vx = x(1-2x) + VC.
VD|C = max q(1-1/4-q) + VX = (3/8)2 + VX
VD|x = max q(1-x-q) + VX = (1-x)2/4 + VX
VC VD|C VC (3/8)2 + 2VC + x(1-2x)
(1-2) VC - (3/8)2 x(1-2x) (1+)/8 - (3/8)2 x(1-2x)
VX VD|C (1-)Vx (1-x)2/4 (1-)(x(1-2x) + /8(1-)) (1-x)2/4
(1-)x(1-2x) + /8 (1-x)2/4
2x2 x + 1/8 9/64 0
(9/4-2)x2 (3-2)x +/8(1-) 0

13